Matt Pocock Skills vs Addy Osmani Agent Skills
Both encode strong engineering practice, but they optimize for different shapes of work. Choose the operating model first, then borrow individual skills carefully.
Matt Pocock Skills
A focused daily engineering loop built around alignment, domain language, feedback, and codebase design.
Read complete guideAddy Osmani Agent Skills
A production engineering system spanning definition, planning, building, verification, review, and launch.
Read complete guideAt a glance
| Dimension | Matt Pocock | Addy Osmani |
|---|---|---|
| Core shape | One expert’s focused, composable daily toolbox | A complete senior-engineering lifecycle |
| Strongest phase | Define and Build | Define through Ship |
| Signature mechanism | Design-tree grilling and shared domain language | Lifecycle routing, quality gates, and behavior evals |
| Architecture | Deep modules, clean seams, active domain modeling | Stable interfaces, ADRs, constraints, and migration planning |
| Production operations | Selective utilities such as Wizard | Security, performance, CI/CD, observability, and launch |
| Platform posture | Claude Code first; portable via skills CLI | Designed and documented across many coding agents |
| Best fit | Experienced individuals and small teams optimizing flow | Teams standardizing a production delivery system |
Choose by situation
Choose Matt when
You want the sharpest requirements interview.
Your codebase needs stronger shared language.
You prefer focused commands over a lifecycle framework.
Claude Code is your primary environment.
Choose Addy when
You need coverage from idea through launch.
Security, performance, and operations must be explicit.
A team needs shared checkpoints and vocabulary.
Skills must work across several agent platforms.
Can you combine them?
Yes, but do not run both complete routing systems at once. Their overlapping TDD, interview, planning, and review skills can compete for invocation and encode different process choices. Pick one collection as the primary operating model, then borrow narrowly scoped skills from the other where the trigger and responsibility are unambiguous.